Neuralyzer is an AI agent harness tool that allows an agent to wipe its own session context and re-run the first message.
The tool solves the problem of context accumulation in agentic loops by giving agents direct control over their session state. Traditional loop patterns either run outside the agent or accumulate messages in the context window, increasing token costs and causing context rot. Neuralyzer adds a single callable tool named `neuralyzer` that, when invoked, clears all user and assistant messages from the session and resends the initial message, allowing the agent to restart fresh without manual intervention or external loop management.
The tool is designed for developers building agentic workflows who want loops controlled by the agent itself rather than external shell commands or interactive commands. It suits projects using harnesses like pi or OpenCode where you want to avoid context window bloat from repeated loop iterations. Unlike `/loop` commands that keep adding to session context, neuralyzer provides a clean restart mechanism. The tool currently supports pi and OpenCode harnesses, with contribution guidelines available for adding support to other harnesses.
Development activity shows active engagement with the extension ecosystem, as evidenced by the project's focus on supporting multiple harness platforms and explicit invitation for pull requests to add new harness integrations. The codebase is written in TypeScript and organized around the core neuralyzer tool implementation with clear documentation for contributors.
